Automatic multi-performance evaluation system for hybrid speech recognition

Abstract

The present invention relates to an automatic multi-performance evaluation system for hybrid speech recognition and, more particularly, to an automatic multi-performance evaluation system for hybrid speech recognition capable of automatically evaluating a processing result of speech recognition and language understanding through a hybrid terminal and a processing result of server speech recognition and language understanding through a cloud server by simultaneously using the hybrid terminal and the cloud server, to thereby output an accurate and fast performance and function verification result.

Description

TECHNICAL FIELD[0001]The present invention relates to an automatic multi-performance evaluation system for hybrid speech recognition and, more particularly, to an automatic multi-performance evaluation system for hybrid speech recognition capable of automatically evaluating a processing result of speech recognition and language understanding through a hybrid terminal and a processing result of server speech recognition and language understanding through a cloud server by simultaneously using the hybrid terminal and the cloud server, to thereby output an accurate and fast performance and function verification result.BACKGROUND ART[0002]Speech recognition refers to technology of analyzing, by a computer, voice of a user input through a microphone, extracting features, recognizing a result of approximating previously input words or sentences as a command, and performing an operation corresponding to the recognized command.[0003]A conventional speech recognition system has separately us...

AI Technical Summary

Benefits of technology

[0033]According to the present invention, an automatic multi-performance evaluation system for hybrid speech recognition can individually or complexly analyze automatic verification of a speech recognition result of a hybrid terminal, automatic verification of a language understanding result of the hybrid terminal, automatic verification of a speech recognition result of a cloud server, automatic verification of a language understanding result of the cloud server, and verification of an arbitration result of the interior of the system and output the analyzed result.

[0034]In addition, functions for simultaneously processing various speech recognition results in a hybrid terminal and a cloud server can be provided so that performance of simple and repeated automation, which has conventionally been individually performed throughout several trials, can be upgraded through a high-efficiency system capable of performing incorporated processing.

[0035]In addition, an automatic evaluation system for hybrid speech recognition for analyzing a recognition process performed in the interior of a hybrid speech recognition platform through logs and automatically evaluating performance of speech recognition can be provided. Simultaneously, reliability of performance evaluation can be greatly raised by providing multiple functions such as matching of screen output, matching of voice assistant output, and recording of test environment information.

[0036]That is, results of screen output verification and voice output verification are provided to a result analyzer and are mapped to evaluation results processed by an integrated result analyzer, so that recognition performance and operation performance can be complexly evaluated.

[0037]For example, usually, since recognition performance and operation performance are separately measured, twice as much time is consumed. However, according to the present invention, evaluation of two items can be automatically performed simultaneously.

[0038]In addition, reliability of vehicle speech recognition performance evaluation can be raised and systematic management for environment information can be performed, by transmitting information about various test environments to a result output unit and mapping the information to multi-performance results, by linkage through a speech recognition evaluation environment recorder between a camera image unit capable of confirming a road surface, weather, and a surrounding situation and a CAN communication analyzer capable of being aware of speed, engine state, air conditioner state, etc. through CAN communication information, in an environment in which a vehicle actually travels.

Problems solved by technology

Accordingly, a conventional speech recognition evaluation system has a difficulty in evaluating hybrid speech recognition.

However, such a vehicle test scheme encounters various problems regarding casting a few hundred native speakers, guiding the native speakers to a test place, managing the native speakers, safety according to a high-speed traveling situation, deterioration of efficiency according to manual recording of the recognition result, much time consumption in refining and analyzing enormous volumes of results, and impossibility of a repetitive test.

Accordingly, it is difficult to perform enough tests to calculate meaningful statistical results and thus technology for solving this problem is needed.

Images